Deep Learning-Based Skin Lesion Multi-class Classification with Global Average Pooling Improvement

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Skin cancer diagnosis relies on visual inspection and biopsy, which have limitations in accuracy and accessibility.
• Developing automated systems for skin lesion classification is crucial for early detection and improved patient outcomes.

Purpose of the Study:

• To propose a novel deep convolutional neural network (DCNN) model for accurate multi-class skin lesion identification and classification.
• To enhance the performance of skin cancer detection using advanced deep learning techniques.

Main Methods:

• A novel DCNN model incorporating global average pooling was developed for skin lesion analysis.
• The HAM10000 dataset, comprising seven skin lesion classes, was utilized for model training and validation.
• Preprocessing involved black hat filtering for artifact removal and resampling for data balancing.

Main Results:

• The proposed DCNN model achieved a highest accuracy of 97.20% in multi-class skin lesion classification.
• Performance was benchmarked against established transfer learning models like ResNet50, VGG-16, MobileNetV2, and DenseNet121.
• Model efficacy was visually confirmed using a graphical user interface (GUI).

Conclusions:

• The developed DCNN model demonstrates superior performance for automated skin lesion classification compared to existing state-of-the-art methods.
• This AI-driven tool shows significant potential as a computer-aided diagnostic aid for dermatologists.
• The findings support the advancement of deep learning applications in medical diagnostics for improved healthcare.